Flower_IDS_IEC104_preprocessing.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "e7edf6ba-bfe6-412b-81b1-dcec09ac476b",
   "metadata": {
    "tags": []
   },
   "source": [
    "# Federated Learning with Flower"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "334a661f-5693-404c-a603-c3e218a9dd59",
   "metadata": {},
   "source": [
    "#### This notebook presents a simple Federated Learning example / application on developing a Machine Learning model as an Intrusion Detection System for defending against Cyberattacks using distributed and private data.\n",
    "#### This example uses Docker containers as independent workers / clients for performing the client model training on the limited and private local client data.\n",
    "#### This example was developed as part of the research conducted at [MetaMind Innovations - MINDS](https://metamind.gr/) in the framework of the [AI4CYBER Horizon Europe](https://metamind.gr/project/ai4cyber-horizon-europe/) project."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4d571908-e42d-4ada-9ccb-8e4e0b150289",
   "metadata": {},
   "source": [
    "### This notebook, as well as the Flower example, were implemented and tested using Python v3.11, Flower v1.5 and TensorFlow v2.14.1"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bb6741b6-b6bd-43e4-8423-55e2d217c118",
   "metadata": {
    "id": "_zFenI3IPpgI",
    "tags": []
   },
   "source": [
    "## Set Up Flower Environment"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "bcfa455c-9791-4d72-babb-251425e8e57a",
   "metadata": {
    "id": "ke7EyuvG0Zyn"
   },
   "outputs": [],
   "source": [
    "# !pip install --quiet --upgrade flwr==1.5.0\n",
    "# !pip install --quiet --upgrade tensorflow-cpu==2.14.1\n",
    "# !pip install --quiet --upgrade scikit-learn\n",
    "# !pip install --quiet --upgrade pandas\n",
    "# !pip install --quiet --upgrade matplotlib\n",
    "# !pip install --quiet --upgrade git+https://github.com/nepiskopos/py7zr.git"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2b2a72ab-d529-4077-80ed-9e719b70c2f0",
   "metadata": {},
   "source": [
    "## Dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3edb0310-f315-4720-9f77-2d01660d5693",
   "metadata": {},
   "source": [
    "This dataset is part of the [IEC 60870-5-104 Intrusion Detection Dataset](http://ieee-dataport.org/documents/iec-60870-5-104-intrusion-detection-dataset) -- DOI: [10.1109/TII.2021.3093905](http://doi.org/10.1109/TII.2021.3093905) -- published by [ITHACA – University of Western Macedonia](http://ithaca.ece.uowm.gr/)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "efb7a1c8-f3fd-49ab-9958-ae6c294b8219",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "...100%, 10 MB, 1659 KB/s, 6 seconds passedArchive '<py7zr.py7zr.SevenZipFile object at 0x7fbadc187eb0>' extracted successfully!\n"
     ]
    }
   ],
   "source": [
    "import errno\n",
    "import os\n",
    "import py7zr\n",
    "import re\n",
    "import shutil\n",
    "import sys\n",
    "import time\n",
    "import urllib\n",
    "\n",
    "\n",
    "def reporthook(count, block_size, total_size):\n",
    "    global start_time\n",
    "    if count == 0:\n",
    "        start_time = time.time()\n",
    "        return\n",
    "    duration = time.time() - start_time\n",
    "    progress_size = int(count * block_size)\n",
    "    speed = int(progress_size / (1024 * duration))\n",
    "    percent = int(count * block_size * 100 / total_size)\n",
    "    sys.stdout.write(\"\\r...%d%%, %d MB, %d KB/s, %d seconds passed\" %\n",
    "                    (percent, progress_size / (1024 * 1024), speed, duration))\n",
    "    sys.stdout.flush()\n",
    "\n",
    "def download(url, filename):\n",
    "    urllib.request.urlretrieve(url, filename, reporthook)\n",
    "\n",
    "def extract(archive, target_dir):\n",
    "    archive = py7zr.SevenZipFile(archive, mode='r')\n",
    "    try:\n",
    "        os.makedirs(target_dir)\n",
    "        \n",
    "        archive.extractall(path=target_dir)\n",
    "        archive.close()\n",
    "        \n",
    "        print(f\"Archive '{archive}' extracted successfully!\")\n",
    "    except OSError as e:\n",
    "        if e.errno == errno.EEXIST:\n",
    "            archive.extractall(path=target_dir)\n",
    "            archive.close()\n",
    "            \n",
    "            print(f\"Archive '{archive}' extracted successfully!\")\n",
    "        else:\n",
    "            print(f\"Archive '{archive}' failed to extract...\")\n",
    "\n",
    "\n",
    "base_data_dir = './datasets'\n",
    "\n",
    "if not os.path.isdir(os.path.join(base_data_dir, 'Balanced_IEC104_Train_Test_CSV_Files')):\n",
    "    filename = 'Balanced_IEC104_Train_Test_CSV_Files.7z'\n",
    "    \n",
    "    if not py7zr.is_7zfile(filename):\n",
    "        url = ('https://zenodo.org/records/7108614/files/Balanced_IEC104_Train_Test_CSV_Files.7z?download=1')\n",
    "        download(url, filename)\n",
    "    \n",
    "    extract(filename, base_data_dir)\n",
    "    \n",
    "    os.remove(os.path.join('.', filename))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "9f976089-9202-4b15-bd15-2997a4d89144",
   "metadata": {},
   "outputs": [],
   "source": [
    "import collections\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from matplotlib import pyplot as plt\n",
    "\n",
    "pd.set_option('future.no_silent_downcasting', True)\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0b4d9f70-d3d0-44ab-baa3-979c88065a37",
   "metadata": {},
   "source": [
    "We must choose which dataset version to facilitate for the experimentation as well as the network flow capturing timeout.\n",
    "\n",
    "We must also choose $3 \\leq N \\leq 5$ distinct federated learning workers.\n",
    "\n",
    "For this example we use the CICFlowMeter dataset with capturing timeout at 180 seconds and 3 workers."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "dec5093d-7510-4fba-b10b-26c82307be00",
   "metadata": {},
   "outputs": [],
   "source": [
    "tp = 'cic'      # Type: dataset type (CICFlowMeter or Custom Python parser), choose between 'cic' and 'custom'\n",
    "tm = 180        # Timeout: data flow timeout, choose a value from the following list [15, 30, 60, 90, 120, 180]\n",
    "n_workers = 3   # N: number of federated workers, choose between 3 and 5"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e0467bfd-f129-464e-ba22-becea8c0993a",
   "metadata": {},
   "source": [
    "Perform some value checks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "5e8ea52b-b893-4251-af46-abef27045ac3",
   "metadata": {},
   "outputs": [],
   "source": [
    "assert tp in ('cic', 'custom'), \"Wrong dataset type, choose between 'cic' and 'custom'\"\n",
    "assert tm in (15, 30, 60, 90, 120, 180), \"Wrong time, choose a value from the following list [15, 30, 60, 90, 120, 180]\"\n",
    "assert n_workers in (3, 4, 5), \"Wrong number of workers, choose between 3 and 5\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "050ffcfa-e371-411a-9172-dde2fb2315c5",
   "metadata": {},
   "outputs": [],
   "source": [
    "path = os.path.join(base_data_dir, 'Balanced_IEC104_Train_Test_CSV_Files/iec104_train_test_csvs')\n",
    "dataset = f'tests_{tp}_{tm}'\n",
    "\n",
    "if 'cic' in dataset:\n",
    "    n = dataset.split('_')[-1]\n",
    "    train_csv = os.path.join(os.path.join(path, dataset), f'train_{tm}_cicflow.csv')\n",
    "    test_csv = os.path.join(os.path.join(path, dataset), f'test_{tm}_cicflow.csv')\n",
    "elif 'custom' in dataset:\n",
    "    n = dataset.split('_')\n",
    "    train_csv = os.path.join(os.path.join(path, dataset), f'train_{tm}_custom_script.csv')\n",
    "    test_csv = os.path.join(os.path.join(path, dataset), f'test_{tm}_custom_script.csv')\n",
    "else:\n",
    "    raise Exception(\"Wrong dataset\")\n",
    "\n",
    "df_train = pd.read_csv(train_csv)\n",
    "df_test = pd.read_csv(train_csv)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1ef57df7-92d2-4cab-a454-e1e04611cb03",
   "metadata": {},
   "source": [
    "Let's perform some data exploration."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "61a1113d-69b5-43ab-986f-e954ab8db2d8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 3588 entries, 0 to 3587\n",
      "Data columns (total 84 columns):\n",
      " #   Column             Non-Null Count  Dtype  \n",
      "---  ------             --------------  -----  \n",
      " 0   Flow ID            3588 non-null   object \n",
      " 1   Src IP             3588 non-null   object \n",
      " 2   Src Port           3588 non-null   int64  \n",
      " 3   Dst IP             3588 non-null   object \n",
      " 4   Dst Port           3588 non-null   int64  \n",
      " 5   Protocol           3588 non-null   int64  \n",
      " 6   Timestamp          3588 non-null   object \n",
      " 7   Flow Duration      3588 non-null   int64  \n",
      " 8   Tot Fwd Pkts       3588 non-null   int64  \n",
      " 9   Tot Bwd Pkts       3588 non-null   int64  \n",
      " 10  TotLen Fwd Pkts    3588 non-null   int64  \n",
      " 11  TotLen Bwd Pkts    3588 non-null   int64  \n",
      " 12  Fwd Pkt Len Max    3588 non-null   int64  \n",
      " 13  Fwd Pkt Len Min    3588 non-null   int64  \n",
      " 14  Fwd Pkt Len Mean   3588 non-null   float64\n",
      " 15  Fwd Pkt Len Std    3588 non-null   float64\n",
      " 16  Bwd Pkt Len Max    3588 non-null   int64  \n",
      " 17  Bwd Pkt Len Min    3588 non-null   int64  \n",
      " 18  Bwd Pkt Len Mean   3588 non-null   float64\n",
      " 19  Bwd Pkt Len Std    3588 non-null   float64\n",
      " 20  Flow Byts/s        3588 non-null   float64\n",
      " 21  Flow Pkts/s        3588 non-null   float64\n",
      " 22  Flow IAT Mean      3588 non-null   float64\n",
      " 23  Flow IAT Std       3588 non-null   float64\n",
      " 24  Flow IAT Max       3588 non-null   int64  \n",
      " 25  Flow IAT Min       3588 non-null   int64  \n",
      " 26  Fwd IAT Tot        3588 non-null   int64  \n",
      " 27  Fwd IAT Mean       3588 non-null   float64\n",
      " 28  Fwd IAT Std        3588 non-null   float64\n",
      " 29  Fwd IAT Max        3588 non-null   int64  \n",
      " 30  Fwd IAT Min        3588 non-null   int64  \n",
      " 31  Bwd IAT Tot        3588 non-null   int64  \n",
      " 32  Bwd IAT Mean       3588 non-null   float64\n",
      " 33  Bwd IAT Std        3588 non-null   float64\n",
      " 34  Bwd IAT Max        3588 non-null   int64  \n",
      " 35  Bwd IAT Min        3588 non-null   int64  \n",
      " 36  Fwd PSH Flags      3588 non-null   int64  \n",
      " 37  Bwd PSH Flags      3588 non-null   int64  \n",
      " 38  Fwd URG Flags      3588 non-null   int64  \n",
      " 39  Bwd URG Flags      3588 non-null   int64  \n",
      " 40  Fwd Header Len     3588 non-null   int64  \n",
      " 41  Bwd Header Len     3588 non-null   int64  \n",
      " 42  Fwd Pkts/s         3588 non-null   float64\n",
      " 43  Bwd Pkts/s         3588 non-null   float64\n",
      " 44  Pkt Len Min        3588 non-null   int64  \n",
      " 45  Pkt Len Max        3588 non-null   int64  \n",
      " 46  Pkt Len Mean       3588 non-null   float64\n",
      " 47  Pkt Len Std        3588 non-null   float64\n",
      " 48  Pkt Len Var        3588 non-null   float64\n",
      " 49  FIN Flag Cnt       3588 non-null   int64  \n",
      " 50  SYN Flag Cnt       3588 non-null   int64  \n",
      " 51  RST Flag Cnt       3588 non-null   int64  \n",
      " 52  PSH Flag Cnt       3588 non-null   int64  \n",
      " 53  ACK Flag Cnt       3588 non-null   int64  \n",
      " 54  URG Flag Cnt       3588 non-null   int64  \n",
      " 55  CWE Flag Count     3588 non-null   int64  \n",
      " 56  ECE Flag Cnt       3588 non-null   int64  \n",
      " 57  Down/Up Ratio      3588 non-null   int64  \n",
      " 58  Pkt Size Avg       3588 non-null   float64\n",
      " 59  Fwd Seg Size Avg   3588 non-null   float64\n",
      " 60  Bwd Seg Size Avg   3588 non-null   float64\n",
      " 61  Fwd Byts/b Avg     3588 non-null   int64  \n",
      " 62  Fwd Pkts/b Avg     3588 non-null   int64  \n",
      " 63  Fwd Blk Rate Avg   3588 non-null   int64  \n",
      " 64  Bwd Byts/b Avg     3588 non-null   int64  \n",
      " 65  Bwd Pkts/b Avg     3588 non-null   int64  \n",
      " 66  Bwd Blk Rate Avg   3588 non-null   int64  \n",
      " 67  Subflow Fwd Pkts   3588 non-null   int64  \n",
      " 68  Subflow Fwd Byts   3588 non-null   int64  \n",
      " 69  Subflow Bwd Pkts   3588 non-null   int64  \n",
      " 70  Subflow Bwd Byts   3588 non-null   int64  \n",
      " 71  Init Fwd Win Byts  3588 non-null   int64  \n",
      " 72  Init Bwd Win Byts  3588 non-null   int64  \n",
      " 73  Fwd Act Data Pkts  3588 non-null   int64  \n",
      " 74  Fwd Seg Size Min   3588 non-null   int64  \n",
      " 75  Active Mean        3588 non-null   float64\n",
      " 76  Active Std         3588 non-null   float64\n",
      " 77  Active Max         3588 non-null   int64  \n",
      " 78  Active Min         3588 non-null   int64  \n",
      " 79  Idle Mean          3588 non-null   float64\n",
      " 80  Idle Std           3588 non-null   float64\n",
      " 81  Idle Max           3588 non-null   int64  \n",
      " 82  Idle Min           3588 non-null   int64  \n",
      " 83  Label              3588 non-null   object \n",
      "dtypes: float64(24), int64(55), object(5)\n",
      "memory usage: 2.3+ MB\n"
     ]
    }
   ],
   "source": [
    "df_train.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "859962a0-9629-495f-946c-25f665f6a408",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 3588 entries, 0 to 3587\n",
      "Data columns (total 84 columns):\n",
      " #   Column             Non-Null Count  Dtype  \n",
      "---  ------             --------------  -----  \n",
      " 0   Flow ID            3588 non-null   object \n",
      " 1   Src IP             3588 non-null   object \n",
      " 2   Src Port           3588 non-null   int64  \n",
      " 3   Dst IP             3588 non-null   object \n",
      " 4   Dst Port           3588 non-null   int64  \n",
      " 5   Protocol           3588 non-null   int64  \n",
      " 6   Timestamp          3588 non-null   object \n",
      " 7   Flow Duration      3588 non-null   int64  \n",
      " 8   Tot Fwd Pkts       3588 non-null   int64  \n",
      " 9   Tot Bwd Pkts       3588 non-null   int64  \n",
      " 10  TotLen Fwd Pkts    3588 non-null   int64  \n",
      " 11  TotLen Bwd Pkts    3588 non-null   int64  \n",
      " 12  Fwd Pkt Len Max    3588 non-null   int64  \n",
      " 13  Fwd Pkt Len Min    3588 non-null   int64  \n",
      " 14  Fwd Pkt Len Mean   3588 non-null   float64\n",
      " 15  Fwd Pkt Len Std    3588 non-null   float64\n",
      " 16  Bwd Pkt Len Max    3588 non-null   int64  \n",
      " 17  Bwd Pkt Len Min    3588 non-null   int64  \n",
      " 18  Bwd Pkt Len Mean   3588 non-null   float64\n",
      " 19  Bwd Pkt Len Std    3588 non-null   float64\n",
      " 20  Flow Byts/s        3588 non-null   float64\n",
      " 21  Flow Pkts/s        3588 non-null   float64\n",
      " 22  Flow IAT Mean      3588 non-null   float64\n",
      " 23  Flow IAT Std       3588 non-null   float64\n",
      " 24  Flow IAT Max       3588 non-null   int64  \n",
      " 25  Flow IAT Min       3588 non-null   int64  \n",
      " 26  Fwd IAT Tot        3588 non-null   int64  \n",
      " 27  Fwd IAT Mean       3588 non-null   float64\n",
      " 28  Fwd IAT Std        3588 non-null   float64\n",
      " 29  Fwd IAT Max        3588 non-null   int64  \n",
      " 30  Fwd IAT Min        3588 non-null   int64  \n",
      " 31  Bwd IAT Tot        3588 non-null   int64  \n",
      " 32  Bwd IAT Mean       3588 non-null   float64\n",
      " 33  Bwd IAT Std        3588 non-null   float64\n",
      " 34  Bwd IAT Max        3588 non-null   int64  \n",
      " 35  Bwd IAT Min        3588 non-null   int64  \n",
      " 36  Fwd PSH Flags      3588 non-null   int64  \n",
      " 37  Bwd PSH Flags      3588 non-null   int64  \n",
      " 38  Fwd URG Flags      3588 non-null   int64  \n",
      " 39  Bwd URG Flags      3588 non-null   int64  \n",
      " 40  Fwd Header Len     3588 non-null   int64  \n",
      " 41  Bwd Header Len     3588 non-null   int64  \n",
      " 42  Fwd Pkts/s         3588 non-null   float64\n",
      " 43  Bwd Pkts/s         3588 non-null   float64\n",
      " 44  Pkt Len Min        3588 non-null   int64  \n",
      " 45  Pkt Len Max        3588 non-null   int64  \n",
      " 46  Pkt Len Mean       3588 non-null   float64\n",
      " 47  Pkt Len Std        3588 non-null   float64\n",
      " 48  Pkt Len Var        3588 non-null   float64\n",
      " 49  FIN Flag Cnt       3588 non-null   int64  \n",
      " 50  SYN Flag Cnt       3588 non-null   int64  \n",
      " 51  RST Flag Cnt       3588 non-null   int64  \n",
      " 52  PSH Flag Cnt       3588 non-null   int64  \n",
      " 53  ACK Flag Cnt       3588 non-null   int64  \n",
      " 54  URG Flag Cnt       3588 non-null   int64  \n",
      " 55  CWE Flag Count     3588 non-null   int64  \n",
      " 56  ECE Flag Cnt       3588 non-null   int64  \n",
      " 57  Down/Up Ratio      3588 non-null   int64  \n",
      " 58  Pkt Size Avg       3588 non-null   float64\n",
      " 59  Fwd Seg Size Avg   3588 non-null   float64\n",
      " 60  Bwd Seg Size Avg   3588 non-null   float64\n",
      " 61  Fwd Byts/b Avg     3588 non-null   int64  \n",
      " 62  Fwd Pkts/b Avg     3588 non-null   int64  \n",
      " 63  Fwd Blk Rate Avg   3588 non-null   int64  \n",
      " 64  Bwd Byts/b Avg     3588 non-null   int64  \n",
      " 65  Bwd Pkts/b Avg     3588 non-null   int64  \n",
      " 66  Bwd Blk Rate Avg   3588 non-null   int64  \n",
      " 67  Subflow Fwd Pkts   3588 non-null   int64  \n",
      " 68  Subflow Fwd Byts   3588 non-null   int64  \n",
      " 69  Subflow Bwd Pkts   3588 non-null   int64  \n",
      " 70  Subflow Bwd Byts   3588 non-null   int64  \n",
      " 71  Init Fwd Win Byts  3588 non-null   int64  \n",
      " 72  Init Bwd Win Byts  3588 non-null   int64  \n",
      " 73  Fwd Act Data Pkts  3588 non-null   int64  \n",
      " 74  Fwd Seg Size Min   3588 non-null   int64  \n",
      " 75  Active Mean        3588 non-null   float64\n",
      " 76  Active Std         3588 non-null   float64\n",
      " 77  Active Max         3588 non-null   int64  \n",
      " 78  Active Min         3588 non-null   int64  \n",
      " 79  Idle Mean          3588 non-null   float64\n",
      " 80  Idle Std           3588 non-null   float64\n",
      " 81  Idle Max           3588 non-null   int64  \n",
      " 82  Idle Min           3588 non-null   int64  \n",
      " 83  Label              3588 non-null   object \n",
      "dtypes: float64(24), int64(55), object(5)\n",
      "memory usage: 2.3+ MB\n"
     ]
    }
   ],
   "source": [
    "df_test.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fcc35f5f-dbc5-4c54-977e-cc1d33a40b4c",
   "metadata": {},
   "source": [
    "As a preprocessing step, we transorm the \"Label\" feature from string to categorical and then to numeric."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "a53d332f-4536-42fe-9a0d-4a91012f8232",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_train['Label'] = df_train['Label'].str.lower()\n",
    "df_test['Label'] = df_test['Label'].str.lower()\n",
    "\n",
    "unique_labels = list(df_train.Label.astype('category').unique())\n",
    "unique_codes = list(df_train.Label.astype('category').cat.codes.unique())\n",
    "\n",
    "mapping = dict(zip(unique_labels, unique_codes))\n",
    "mapping_inv = dict(zip(unique_codes, unique_labels))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "a7e4857d-1a9d-4e69-b0d5-02a6c3840875",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'normal': 11,\n",
       " 'c_ci_na_1': 0,\n",
       " 'c_ci_na_1_dos': 1,\n",
       " 'c_rd_na_1': 2,\n",
       " 'c_rd_na_1_dos': 3,\n",
       " 'c_rp_na_1': 4,\n",
       " 'c_rp_na_1_dos': 5,\n",
       " 'c_sc_na_1': 6,\n",
       " 'c_sc_na_1_dos': 7,\n",
       " 'c_se_na_1': 8,\n",
       " 'c_se_na_1_dos': 9,\n",
       " 'm_sp_na_1_dos': 10}"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mapping"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "6742ef54-dfde-4163-a5a8-ce77ce96d0e9",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_train['Label'] = df_train['Label'].astype('category').cat.rename_categories(mapping)\n",
    "df_test['Label'] = df_test['Label'].astype('category').cat.rename_categories(mapping)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7edd853c-a1c3-404a-93d5-6a7a82dcb377",
   "metadata": {},
   "source": [
    "Now, we remove the features which can not contribute in training a ML model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "d6724255-5c6e-4e88-8712-ed65cc2816df",
   "metadata": {},
   "outputs": [],
   "source": [
    "if tp == 'cic':\n",
    "    train = df_train.iloc[:,7:]\n",
    "    test = df_test.iloc[:,7:]\n",
    "else:\n",
    "    train = df_train\n",
    "    test = df_test"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "21e7b4f4-891e-452f-ac11-6ca09d1b3a79",
   "metadata": {},
   "source": [
    "We rename the \"Label\" column to \"y\"."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "0c909bee-4500-4753-90dc-80c37695c0f5",
   "metadata": {},
   "outputs": [],
   "source": [
    "train.rename(columns={\"Label\": \"y\"}, inplace=True, errors=\"raise\")\n",
    "test.rename(columns={\"Label\": \"y\"}, inplace=True, errors=\"raise\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "daf2cc94-839b-49d1-a588-b38f7bafd1dd",
   "metadata": {},
   "source": [
    "We drop Inf and NaN values both in the training and in test feature data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "cc5d3627-cf4f-4882-987c-a6b74fda51f9",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Replace Inf values with NaN\n",
    "train.replace([np.inf, -np.inf], np.nan, inplace=True)\n",
    "test.replace([np.inf, -np.inf], np.nan, inplace=True)\n",
    "\n",
    "# Drop rows with NaN values\n",
    "train.dropna(inplace=True)\n",
    "test.dropna(inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ae1c91e7-2bd0-4d4c-8498-edbf3bcd59bd",
   "metadata": {},
   "source": [
    "We store the modified datasets into new local CSV files for future use and model evaluation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "d919d458-a25f-4da3-af63-1b0caed8f5f7",
   "metadata": {},
   "outputs": [],
   "source": [
    "directory = os.path.join(base_data_dir, 'federated_datasets')\n",
    "\n",
    "try:\n",
    "    os.makedirs(directory)\n",
    "    train.to_csv(os.path.join(directory, 'train_data.csv'), index=False)\n",
    "    test.to_csv(os.path.join(directory, 'test_data.csv'), index=False)\n",
    "except OSError as e:\n",
    "    if e.errno != errno.EEXIST:\n",
    "        raise\n",
    "    else:   \n",
    "        train.to_csv(os.path.join(directory, 'train_data.csv'), index=False)\n",
    "        test.to_csv(os.path.join(directory, 'test_data.csv'), index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "695bb658-206d-4cbd-954a-edb49b44d11b",
   "metadata": {},
   "source": [
    "Each worker will be assigned a subset of the dataset with $\\lfloor {M \\over N} \\rfloor$ data points.\n",
    "\n",
    "Each worker data is stored as a seperate CSV file with the proper naming scheme."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "cff97f6f-42ba-4525-9157-4693c6f3663f",
   "metadata": {},
   "outputs": [],
   "source": [
    "n_samples = int(train.shape[0] / n_workers)\n",
    "\n",
    "assert type(n_workers) == int, \"Non-int number of workers\"\n",
    "assert n_workers >= 3 and n_workers <= df_train.shape[0], \"At least 3 workers and at most as many workers as the number of samples are allowed\"\n",
    "assert n_samples > 0, \"Each worker must be assigned at least one data point\""
   ]
  },
  {
   "cell_type": "markdown",
   "id": "35dc41bd-07c1-4176-86c6-1737061ef93e",
   "metadata": {},
   "source": [
    "Each worker's training data is stored as a seperate CSV file with the proper naming and data format schemes."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "f7ff7dcb-9ae2-4c44-9458-efc3616d2cde",
   "metadata": {},
   "outputs": [],
   "source": [
    "client_data = []\n",
    "train_copy = train.copy()\n",
    "\n",
    "for i in range(n_workers):\n",
    "    sample = train_copy.sample(n=n_samples)\n",
    "    sample.to_csv(os.path.join(directory, f'client_train_data_{i+1}.csv'), index=False)\n",
    "    train_copy.drop(index=sample.index, inplace=True)\n",
    "    client_data.append(sample)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d3041faf-16e5-4c15-86cf-743821a7cb4b",
   "metadata": {},
   "source": [
    "Let's check the number of points in each straggler's training data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "c3346012-6bf9-4879-80b9-f46f890dfb72",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Worker 1 training data contains 1196 points\n",
      "Worker 2 training data contains 1196 points\n",
      "Worker 3 training data contains 1196 points\n"
     ]
    }
   ],
   "source": [
    "for i in range(n_workers):\n",
    "    print(f\"Worker {i+1} training data contains {len(client_data[i])} points\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "24226028-58fd-4da0-a83b-571e779910fa",
   "metadata": {},
   "source": [
    "Let's perform some statistical analysis on the difference in the number of samples for each digit in each client dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "726964a7-87c4-4001-ac17-2fcaa73dfe3c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 2000x700 with 3 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Number of examples per layer for a sample of clients\n",
    "fig = plt.figure(figsize=(20, 7))\n",
    "fig.suptitle('Label Counts for a Sample of Worker Data')\n",
    "fig.tight_layout()\n",
    "\n",
    "for i in range(n_workers):\n",
    "    m = 0\n",
    "    plot_data = collections.defaultdict(list)\n",
    "    sample = client_data[i]\n",
    "    for label in sample['y']:\n",
    "        # Append counts individually per label to make plots\n",
    "        # more colorful instead of one color per plot.\n",
    "        plot_data[label].append(label)\n",
    "        m = max(m, len(plot_data[label]))\n",
    "    \n",
    "    n_cols = n_workers if n_workers < 5 else 5\n",
    "    xlim = [0, m+(20-m%20)]\n",
    "    ylim = [min(unique_codes)-1, max(unique_codes)+1]\n",
    "    yticks = list(range(min(unique_codes), max(unique_codes)+1))\n",
    "    yticks_labels = [mapping_inv[k] for k in range(0, max(unique_codes)+1)]\n",
    "    \n",
    "    plt.subplot(int(n_workers / 5)+1, n_cols, i+1)\n",
    "    plt.subplots_adjust(wspace=0.3)\n",
    "    plt.title('Worker {}'.format(i+1))\n",
    "    plt.xlabel('#points')\n",
    "    plt.xlim(xlim)\n",
    "    # plt.ylabel('Label')\n",
    "    plt.ylim(ylim)\n",
    "    plt.yticks(yticks, labels=yticks_labels)\n",
    "    \n",
    "    # plot values on top of bars\n",
    "    for key in plot_data:\n",
    "        if len(plot_data[key]) > 0:\n",
    "            plt.text(len(plot_data[key])+4, int(key)-0.1, str(len(plot_data[key])), ha='center')\n",
    "    \n",
    "    for j in range(min(unique_codes),max(unique_codes)+1):\n",
    "        plt.hist(\n",
    "            plot_data[j],\n",
    "            density=False,\n",
    "            bins=[k-0.5 for k in range(min(unique_codes),max(unique_codes)+2)],\n",
    "            orientation='horizontal'\n",
    "        )"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6bd712f9-4e9f-45a4-bb27-935e0eab0e7a",
   "metadata": {
    "id": "8Lq8r5uaT2rB",
    "tags": []
   },
   "source": [
    "### Launch $N=3$ Flower FL workers as Docker Containers locally using the **docker-compose.yaml**\n",
    "\n",
    "```\n",
    "docker-compose -f ./docker-compose.yaml up --scale flwr_c=3\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "81051f3d-f18c-4ff6-bdad-5061ecdfda63",
   "metadata": {},
   "source": [
    "## Centralized model"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b08c9ffe-ed35-4138-802a-a4d235564508",
   "metadata": {},
   "source": [
    "Let's train a centralized IDS model to use as a baseline model for evaluation reference."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "e8024ec9-e29b-4ab6-bb74-6bd2415edc45",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-04-29 16:46:54.134814: I tensorflow/core/util/port.cc:111] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.\n",
      "2024-04-29 16:46:54.152519: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.\n",
      "To enable the following instructions: AVX2 AVX_VNNI FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "from tensorflow.keras.callbacks import EarlyStopping\n",
    "from tensorflow.keras.utils import to_categorical\n",
    "\n",
    "from sklearn.metrics import f1_score\n",
    "from sklearn.preprocessing import StandardScaler"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6276e4a1-b243-437b-afa6-f593e0ee884f",
   "metadata": {},
   "source": [
    "We have to scale the data to avoid huge value deviations."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "127b667d-77f6-482b-aae5-d800adca0d3e",
   "metadata": {},
   "outputs": [],
   "source": [
    "scaler = StandardScaler()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "c7b08368-e074-4150-8515-fe1bd8c7a1bd",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train = scaler.fit_transform(train.drop(['y'], axis=1).to_numpy())\n",
    "y_train = train['y'].to_numpy()\n",
    "X_test = scaler.transform(test.drop(['y'], axis=1).to_numpy())\n",
    "y_test = test['y'].to_numpy()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ee7d0089-c292-43bd-85ac-91c9621642b1",
   "metadata": {},
   "source": [
    "Then, we are better off using one-hot-vectors than scalars for the label representation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "d5753b11-863c-42be-a424-c4fd2012ee2b",
   "metadata": {},
   "outputs": [],
   "source": [
    "y_train_cat = to_categorical(y_train)\n",
    "y_test_cat = to_categorical(y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "85d0f318-4542-4b2f-bf8b-1e123e9a26f4",
   "metadata": {},
   "source": [
    "We define a MLP model with one hidden layer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "529f5022-2c19-4d4e-950a-2bb014b87a0f",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = tf.keras.models.Sequential([\n",
    "    tf.keras.layers.InputLayer(input_shape=(X_train.shape[1],)),\n",
    "    tf.keras.layers.Dense(units=50, activation='relu'),\n",
    "    tf.keras.layers.Dropout(0.2),\n",
    "    tf.keras.layers.Dense(units=y_train_cat.shape[1], activation='softmax')\n",
    "])\n",
    "\n",
    "model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "954ad825-66c2-4d04-a2c1-a7c5a9161478",
   "metadata": {},
   "source": [
    "We use early stopping monitoring the validation loss to avoid overfitting."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "b7baafac-cbb7-425f-b70f-55f6bb2302a6",
   "metadata": {},
   "outputs": [],
   "source": [
    "early_stop = EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=20)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "545fff1a-5ce8-45ad-8729-b87ddee56e66",
   "metadata": {},
   "source": [
    "Let's train the model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "678d1c40-6a87-4906-8996-fa38ad442695",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/300\n",
      "57/57 [==============================] - 1s 4ms/step - loss: 2.2518 - accuracy: 0.2475 - val_loss: 1.8757 - val_accuracy: 0.3487\n",
      "Epoch 2/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 1.8068 - accuracy: 0.3579 - val_loss: 1.6816 - val_accuracy: 0.3701\n",
      "Epoch 3/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 1.6533 - accuracy: 0.3899 - val_loss: 1.5405 - val_accuracy: 0.4799\n",
      "Epoch 4/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 1.5396 - accuracy: 0.4378 - val_loss: 1.4386 - val_accuracy: 0.5106\n",
      "Epoch 5/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 1.4541 - accuracy: 0.4440 - val_loss: 1.3612 - val_accuracy: 0.5295\n",
      "Epoch 6/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 1.3838 - accuracy: 0.4827 - val_loss: 1.3000 - val_accuracy: 0.5368\n",
      "Epoch 7/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 1.3332 - accuracy: 0.4958 - val_loss: 1.2537 - val_accuracy: 0.5424\n",
      "Epoch 8/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 1.2988 - accuracy: 0.4847 - val_loss: 1.2131 - val_accuracy: 0.5390\n",
      "Epoch 9/300\n",
      "57/57 [==============================] - 0s 5ms/step - loss: 1.2510 - accuracy: 0.5095 - val_loss: 1.1786 - val_accuracy: 0.5452\n",
      "Epoch 10/300\n",
      "57/57 [==============================] - 0s 4ms/step - loss: 1.2372 - accuracy: 0.5014 - val_loss: 1.1520 - val_accuracy: 0.5424\n",
      "Epoch 11/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 1.2012 - accuracy: 0.5209 - val_loss: 1.1284 - val_accuracy: 0.5446\n",
      "Epoch 12/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 1.1703 - accuracy: 0.5181 - val_loss: 1.1060 - val_accuracy: 0.5443\n",
      "Epoch 13/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 1.1558 - accuracy: 0.5134 - val_loss: 1.0866 - val_accuracy: 0.5493\n",
      "Epoch 14/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 1.1335 - accuracy: 0.5279 - val_loss: 1.0707 - val_accuracy: 0.5513\n",
      "Epoch 15/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 1.1200 - accuracy: 0.5170 - val_loss: 1.0564 - val_accuracy: 0.5452\n",
      "Epoch 16/300\n",
      "57/57 [==============================] - 0s 4ms/step - loss: 1.1067 - accuracy: 0.5242 - val_loss: 1.0444 - val_accuracy: 0.5485\n",
      "Epoch 17/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 1.0934 - accuracy: 0.5206 - val_loss: 1.0315 - val_accuracy: 0.5516\n",
      "Epoch 18/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 1.0840 - accuracy: 0.5295 - val_loss: 1.0213 - val_accuracy: 0.5535\n",
      "Epoch 19/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 1.0690 - accuracy: 0.5337 - val_loss: 1.0118 - val_accuracy: 0.5596\n",
      "Epoch 20/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 1.0547 - accuracy: 0.5273 - val_loss: 1.0030 - val_accuracy: 0.5577\n",
      "Epoch 21/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 1.0521 - accuracy: 0.5265 - val_loss: 0.9933 - val_accuracy: 0.5610\n",
      "Epoch 22/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 1.0391 - accuracy: 0.5393 - val_loss: 0.9858 - val_accuracy: 0.5616\n",
      "Epoch 23/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 1.0270 - accuracy: 0.5435 - val_loss: 0.9794 - val_accuracy: 0.5574\n",
      "Epoch 24/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 1.0200 - accuracy: 0.5287 - val_loss: 0.9728 - val_accuracy: 0.5491\n",
      "Epoch 25/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 1.0186 - accuracy: 0.5438 - val_loss: 0.9650 - val_accuracy: 0.5624\n",
      "Epoch 26/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 1.0118 - accuracy: 0.5424 - val_loss: 0.9588 - val_accuracy: 0.5560\n",
      "Epoch 27/300\n",
      "57/57 [==============================] - 0s 4ms/step - loss: 1.0032 - accuracy: 0.5429 - val_loss: 0.9524 - val_accuracy: 0.5630\n",
      "Epoch 28/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.9936 - accuracy: 0.5421 - val_loss: 0.9473 - val_accuracy: 0.5655\n",
      "Epoch 29/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.9900 - accuracy: 0.5491 - val_loss: 0.9421 - val_accuracy: 0.5622\n",
      "Epoch 30/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.9819 - accuracy: 0.5535 - val_loss: 0.9367 - val_accuracy: 0.5686\n",
      "Epoch 31/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.9763 - accuracy: 0.5535 - val_loss: 0.9300 - val_accuracy: 0.5686\n",
      "Epoch 32/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.9718 - accuracy: 0.5435 - val_loss: 0.9264 - val_accuracy: 0.5608\n",
      "Epoch 33/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.9730 - accuracy: 0.5446 - val_loss: 0.9203 - val_accuracy: 0.5792\n",
      "Epoch 34/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.9603 - accuracy: 0.5543 - val_loss: 0.9164 - val_accuracy: 0.5633\n",
      "Epoch 35/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.9646 - accuracy: 0.5399 - val_loss: 0.9123 - val_accuracy: 0.5722\n",
      "Epoch 36/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.9522 - accuracy: 0.5502 - val_loss: 0.9076 - val_accuracy: 0.5758\n",
      "Epoch 37/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.9509 - accuracy: 0.5527 - val_loss: 0.9028 - val_accuracy: 0.5744\n",
      "Epoch 38/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.9427 - accuracy: 0.5412 - val_loss: 0.8999 - val_accuracy: 0.5744\n",
      "Epoch 39/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.9380 - accuracy: 0.5432 - val_loss: 0.8954 - val_accuracy: 0.5702\n",
      "Epoch 40/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.9355 - accuracy: 0.5513 - val_loss: 0.8916 - val_accuracy: 0.5775\n",
      "Epoch 41/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.9318 - accuracy: 0.5549 - val_loss: 0.8881 - val_accuracy: 0.5825\n",
      "Epoch 42/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.9217 - accuracy: 0.5644 - val_loss: 0.8839 - val_accuracy: 0.5747\n",
      "Epoch 43/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.9271 - accuracy: 0.5582 - val_loss: 0.8824 - val_accuracy: 0.5775\n",
      "Epoch 44/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.9204 - accuracy: 0.5655 - val_loss: 0.8799 - val_accuracy: 0.5739\n",
      "Epoch 45/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.9234 - accuracy: 0.5602 - val_loss: 0.8761 - val_accuracy: 0.5847\n",
      "Epoch 46/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.9229 - accuracy: 0.5535 - val_loss: 0.8739 - val_accuracy: 0.5811\n",
      "Epoch 47/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.9136 - accuracy: 0.5585 - val_loss: 0.8721 - val_accuracy: 0.5794\n",
      "Epoch 48/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.9124 - accuracy: 0.5700 - val_loss: 0.8685 - val_accuracy: 0.5833\n",
      "Epoch 49/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.9083 - accuracy: 0.5647 - val_loss: 0.8663 - val_accuracy: 0.5897\n",
      "Epoch 50/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.9045 - accuracy: 0.5588 - val_loss: 0.8651 - val_accuracy: 0.5822\n",
      "Epoch 51/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.9062 - accuracy: 0.5694 - val_loss: 0.8620 - val_accuracy: 0.5884\n",
      "Epoch 52/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8983 - accuracy: 0.5594 - val_loss: 0.8600 - val_accuracy: 0.5856\n",
      "Epoch 53/300\n",
      "57/57 [==============================] - 0s 4ms/step - loss: 0.8954 - accuracy: 0.5638 - val_loss: 0.8577 - val_accuracy: 0.5814\n",
      "Epoch 54/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8946 - accuracy: 0.5610 - val_loss: 0.8560 - val_accuracy: 0.5928\n",
      "Epoch 55/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8901 - accuracy: 0.5683 - val_loss: 0.8549 - val_accuracy: 0.5909\n",
      "Epoch 56/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8919 - accuracy: 0.5688 - val_loss: 0.8530 - val_accuracy: 0.5945\n",
      "Epoch 57/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8851 - accuracy: 0.5686 - val_loss: 0.8514 - val_accuracy: 0.5864\n",
      "Epoch 58/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8822 - accuracy: 0.5716 - val_loss: 0.8500 - val_accuracy: 0.5839\n",
      "Epoch 59/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8807 - accuracy: 0.5747 - val_loss: 0.8483 - val_accuracy: 0.5920\n",
      "Epoch 60/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8872 - accuracy: 0.5649 - val_loss: 0.8475 - val_accuracy: 0.5923\n",
      "Epoch 61/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8849 - accuracy: 0.5680 - val_loss: 0.8482 - val_accuracy: 0.5900\n",
      "Epoch 62/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8745 - accuracy: 0.5733 - val_loss: 0.8457 - val_accuracy: 0.5895\n",
      "Epoch 63/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8862 - accuracy: 0.5674 - val_loss: 0.8434 - val_accuracy: 0.5870\n",
      "Epoch 64/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8757 - accuracy: 0.5766 - val_loss: 0.8422 - val_accuracy: 0.5856\n",
      "Epoch 65/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8808 - accuracy: 0.5736 - val_loss: 0.8409 - val_accuracy: 0.5903\n",
      "Epoch 66/300\n",
      "57/57 [==============================] - 0s 4ms/step - loss: 0.8755 - accuracy: 0.5758 - val_loss: 0.8393 - val_accuracy: 0.5934\n",
      "Epoch 67/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8736 - accuracy: 0.5778 - val_loss: 0.8394 - val_accuracy: 0.5920\n",
      "Epoch 68/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8757 - accuracy: 0.5778 - val_loss: 0.8378 - val_accuracy: 0.6006\n",
      "Epoch 69/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8766 - accuracy: 0.5741 - val_loss: 0.8357 - val_accuracy: 0.5881\n",
      "Epoch 70/300\n",
      "57/57 [==============================] - 0s 4ms/step - loss: 0.8771 - accuracy: 0.5800 - val_loss: 0.8348 - val_accuracy: 0.5967\n",
      "Epoch 71/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8762 - accuracy: 0.5744 - val_loss: 0.8349 - val_accuracy: 0.5987\n",
      "Epoch 72/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8706 - accuracy: 0.5755 - val_loss: 0.8341 - val_accuracy: 0.5925\n",
      "Epoch 73/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8725 - accuracy: 0.5792 - val_loss: 0.8327 - val_accuracy: 0.5978\n",
      "Epoch 74/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8646 - accuracy: 0.5764 - val_loss: 0.8311 - val_accuracy: 0.5984\n",
      "Epoch 75/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8673 - accuracy: 0.5833 - val_loss: 0.8307 - val_accuracy: 0.6012\n",
      "Epoch 76/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8696 - accuracy: 0.5797 - val_loss: 0.8293 - val_accuracy: 0.5953\n",
      "Epoch 77/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8576 - accuracy: 0.5805 - val_loss: 0.8266 - val_accuracy: 0.5967\n",
      "Epoch 78/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8647 - accuracy: 0.5953 - val_loss: 0.8251 - val_accuracy: 0.5995\n",
      "Epoch 79/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8660 - accuracy: 0.5747 - val_loss: 0.8250 - val_accuracy: 0.6001\n",
      "Epoch 80/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8662 - accuracy: 0.5775 - val_loss: 0.8255 - val_accuracy: 0.5925\n",
      "Epoch 81/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8599 - accuracy: 0.5708 - val_loss: 0.8260 - val_accuracy: 0.5989\n",
      "Epoch 82/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8616 - accuracy: 0.5831 - val_loss: 0.8250 - val_accuracy: 0.5909\n",
      "Epoch 83/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8525 - accuracy: 0.5700 - val_loss: 0.8236 - val_accuracy: 0.6026\n",
      "Epoch 84/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8591 - accuracy: 0.5794 - val_loss: 0.8221 - val_accuracy: 0.6045\n",
      "Epoch 85/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8529 - accuracy: 0.5775 - val_loss: 0.8209 - val_accuracy: 0.6065\n",
      "Epoch 86/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8533 - accuracy: 0.5811 - val_loss: 0.8203 - val_accuracy: 0.6040\n",
      "Epoch 87/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8655 - accuracy: 0.5769 - val_loss: 0.8204 - val_accuracy: 0.6031\n",
      "Epoch 88/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8571 - accuracy: 0.5794 - val_loss: 0.8189 - val_accuracy: 0.5897\n",
      "Epoch 89/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8545 - accuracy: 0.5822 - val_loss: 0.8195 - val_accuracy: 0.6051\n",
      "Epoch 90/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8548 - accuracy: 0.5817 - val_loss: 0.8181 - val_accuracy: 0.6017\n",
      "Epoch 91/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8468 - accuracy: 0.5758 - val_loss: 0.8165 - val_accuracy: 0.5998\n",
      "Epoch 92/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8550 - accuracy: 0.5814 - val_loss: 0.8165 - val_accuracy: 0.6065\n",
      "Epoch 93/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8596 - accuracy: 0.5753 - val_loss: 0.8164 - val_accuracy: 0.6026\n",
      "Epoch 94/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8485 - accuracy: 0.5780 - val_loss: 0.8162 - val_accuracy: 0.5981\n",
      "Epoch 95/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8492 - accuracy: 0.5811 - val_loss: 0.8159 - val_accuracy: 0.5953\n",
      "Epoch 96/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8490 - accuracy: 0.5814 - val_loss: 0.8171 - val_accuracy: 0.5987\n",
      "Epoch 97/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8474 - accuracy: 0.5825 - val_loss: 0.8144 - val_accuracy: 0.6034\n",
      "Epoch 98/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8406 - accuracy: 0.5839 - val_loss: 0.8134 - val_accuracy: 0.5973\n",
      "Epoch 99/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8472 - accuracy: 0.5775 - val_loss: 0.8112 - val_accuracy: 0.6014\n",
      "Epoch 100/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8480 - accuracy: 0.5833 - val_loss: 0.8103 - val_accuracy: 0.6051\n",
      "Epoch 101/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8389 - accuracy: 0.5803 - val_loss: 0.8117 - val_accuracy: 0.6048\n",
      "Epoch 102/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8451 - accuracy: 0.5836 - val_loss: 0.8108 - val_accuracy: 0.6073\n",
      "Epoch 103/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8448 - accuracy: 0.5875 - val_loss: 0.8113 - val_accuracy: 0.5984\n",
      "Epoch 104/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8383 - accuracy: 0.5808 - val_loss: 0.8112 - val_accuracy: 0.6059\n",
      "Epoch 105/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8486 - accuracy: 0.5931 - val_loss: 0.8109 - val_accuracy: 0.6037\n",
      "Epoch 106/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8323 - accuracy: 0.5836 - val_loss: 0.8090 - val_accuracy: 0.6062\n",
      "Epoch 107/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8420 - accuracy: 0.5875 - val_loss: 0.8075 - val_accuracy: 0.6059\n",
      "Epoch 108/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8353 - accuracy: 0.5864 - val_loss: 0.8088 - val_accuracy: 0.6037\n",
      "Epoch 109/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8511 - accuracy: 0.5847 - val_loss: 0.8085 - val_accuracy: 0.6079\n",
      "Epoch 110/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8431 - accuracy: 0.5808 - val_loss: 0.8068 - val_accuracy: 0.6070\n",
      "Epoch 111/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8391 - accuracy: 0.5906 - val_loss: 0.8073 - val_accuracy: 0.6095\n",
      "Epoch 112/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8426 - accuracy: 0.5884 - val_loss: 0.8073 - val_accuracy: 0.6090\n",
      "Epoch 113/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8420 - accuracy: 0.5828 - val_loss: 0.8076 - val_accuracy: 0.5959\n",
      "Epoch 114/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8315 - accuracy: 0.5822 - val_loss: 0.8051 - val_accuracy: 0.6042\n",
      "Epoch 115/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8373 - accuracy: 0.5881 - val_loss: 0.8056 - val_accuracy: 0.6042\n",
      "Epoch 116/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8351 - accuracy: 0.5858 - val_loss: 0.8038 - val_accuracy: 0.6034\n",
      "Epoch 117/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8310 - accuracy: 0.5797 - val_loss: 0.8034 - val_accuracy: 0.6079\n",
      "Epoch 118/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8423 - accuracy: 0.5856 - val_loss: 0.8045 - val_accuracy: 0.6120\n",
      "Epoch 119/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8324 - accuracy: 0.5889 - val_loss: 0.8029 - val_accuracy: 0.6079\n",
      "Epoch 120/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8247 - accuracy: 0.5792 - val_loss: 0.8013 - val_accuracy: 0.6118\n",
      "Epoch 121/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8266 - accuracy: 0.5928 - val_loss: 0.8005 - val_accuracy: 0.6070\n",
      "Epoch 122/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8342 - accuracy: 0.5906 - val_loss: 0.8026 - val_accuracy: 0.6129\n",
      "Epoch 123/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8278 - accuracy: 0.5903 - val_loss: 0.8022 - val_accuracy: 0.6090\n",
      "Epoch 124/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8396 - accuracy: 0.5828 - val_loss: 0.8017 - val_accuracy: 0.6020\n",
      "Epoch 125/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8328 - accuracy: 0.5831 - val_loss: 0.7993 - val_accuracy: 0.6073\n",
      "Epoch 126/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8259 - accuracy: 0.5928 - val_loss: 0.8011 - val_accuracy: 0.6109\n",
      "Epoch 127/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8274 - accuracy: 0.5886 - val_loss: 0.7998 - val_accuracy: 0.6056\n",
      "Epoch 128/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8380 - accuracy: 0.5900 - val_loss: 0.7967 - val_accuracy: 0.6162\n",
      "Epoch 129/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8417 - accuracy: 0.5897 - val_loss: 0.7969 - val_accuracy: 0.6126\n",
      "Epoch 130/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8307 - accuracy: 0.5906 - val_loss: 0.7984 - val_accuracy: 0.6031\n",
      "Epoch 131/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8284 - accuracy: 0.5822 - val_loss: 0.7984 - val_accuracy: 0.6081\n",
      "Epoch 132/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8294 - accuracy: 0.5870 - val_loss: 0.7967 - val_accuracy: 0.6090\n",
      "Epoch 133/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8349 - accuracy: 0.5870 - val_loss: 0.7960 - val_accuracy: 0.6079\n",
      "Epoch 134/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8291 - accuracy: 0.5928 - val_loss: 0.7954 - val_accuracy: 0.6126\n",
      "Epoch 135/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8248 - accuracy: 0.5881 - val_loss: 0.7952 - val_accuracy: 0.6120\n",
      "Epoch 136/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8370 - accuracy: 0.5889 - val_loss: 0.7954 - val_accuracy: 0.6148\n",
      "Epoch 137/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8308 - accuracy: 0.5975 - val_loss: 0.7959 - val_accuracy: 0.6056\n",
      "Epoch 138/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8268 - accuracy: 0.5906 - val_loss: 0.7968 - val_accuracy: 0.6098\n",
      "Epoch 139/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8235 - accuracy: 0.5956 - val_loss: 0.7949 - val_accuracy: 0.6176\n",
      "Epoch 140/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8321 - accuracy: 0.5906 - val_loss: 0.7955 - val_accuracy: 0.6145\n",
      "Epoch 141/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8206 - accuracy: 0.5956 - val_loss: 0.7947 - val_accuracy: 0.6120\n",
      "Epoch 142/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8304 - accuracy: 0.5911 - val_loss: 0.7933 - val_accuracy: 0.6081\n",
      "Epoch 143/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8158 - accuracy: 0.5948 - val_loss: 0.7923 - val_accuracy: 0.6134\n",
      "Epoch 144/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8247 - accuracy: 0.5923 - val_loss: 0.7932 - val_accuracy: 0.6171\n",
      "Epoch 145/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8225 - accuracy: 0.5923 - val_loss: 0.7931 - val_accuracy: 0.6067\n",
      "Epoch 146/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8278 - accuracy: 0.5923 - val_loss: 0.7912 - val_accuracy: 0.6190\n",
      "Epoch 147/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8248 - accuracy: 0.5931 - val_loss: 0.7910 - val_accuracy: 0.6031\n",
      "Epoch 148/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8277 - accuracy: 0.5892 - val_loss: 0.7911 - val_accuracy: 0.6098\n",
      "Epoch 149/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8289 - accuracy: 0.5889 - val_loss: 0.7901 - val_accuracy: 0.6171\n",
      "Epoch 150/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8219 - accuracy: 0.6026 - val_loss: 0.7912 - val_accuracy: 0.6137\n",
      "Epoch 151/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8289 - accuracy: 0.5900 - val_loss: 0.7888 - val_accuracy: 0.6154\n",
      "Epoch 152/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8254 - accuracy: 0.6020 - val_loss: 0.7877 - val_accuracy: 0.6129\n",
      "Epoch 153/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8242 - accuracy: 0.5942 - val_loss: 0.7906 - val_accuracy: 0.6093\n",
      "Epoch 154/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8300 - accuracy: 0.5858 - val_loss: 0.7895 - val_accuracy: 0.6165\n",
      "Epoch 155/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8265 - accuracy: 0.6031 - val_loss: 0.7882 - val_accuracy: 0.6159\n",
      "Epoch 156/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8236 - accuracy: 0.5872 - val_loss: 0.7877 - val_accuracy: 0.6171\n",
      "Epoch 157/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8313 - accuracy: 0.5939 - val_loss: 0.7876 - val_accuracy: 0.6040\n",
      "Epoch 158/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8285 - accuracy: 0.6031 - val_loss: 0.7879 - val_accuracy: 0.6048\n",
      "Epoch 159/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8303 - accuracy: 0.5942 - val_loss: 0.7864 - val_accuracy: 0.6159\n",
      "Epoch 160/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8197 - accuracy: 0.5978 - val_loss: 0.7864 - val_accuracy: 0.6173\n",
      "Epoch 161/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8218 - accuracy: 0.5964 - val_loss: 0.7856 - val_accuracy: 0.6196\n",
      "Epoch 162/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8215 - accuracy: 0.6014 - val_loss: 0.7849 - val_accuracy: 0.6201\n",
      "Epoch 163/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8109 - accuracy: 0.5948 - val_loss: 0.7845 - val_accuracy: 0.6171\n",
      "Epoch 164/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8222 - accuracy: 0.5889 - val_loss: 0.7857 - val_accuracy: 0.6162\n",
      "Epoch 165/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8207 - accuracy: 0.5953 - val_loss: 0.7829 - val_accuracy: 0.6196\n",
      "Epoch 166/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8190 - accuracy: 0.5978 - val_loss: 0.7835 - val_accuracy: 0.6196\n",
      "Epoch 167/300\n",
      "57/57 [==============================] - 0s 4ms/step - loss: 0.8161 - accuracy: 0.5959 - val_loss: 0.7830 - val_accuracy: 0.6196\n",
      "Epoch 168/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8064 - accuracy: 0.5998 - val_loss: 0.7826 - val_accuracy: 0.6112\n",
      "Epoch 169/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8229 - accuracy: 0.5964 - val_loss: 0.7828 - val_accuracy: 0.6093\n",
      "Epoch 170/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8230 - accuracy: 0.5878 - val_loss: 0.7820 - val_accuracy: 0.6123\n",
      "Epoch 171/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8012 - accuracy: 0.5970 - val_loss: 0.7810 - val_accuracy: 0.6106\n",
      "Epoch 172/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8140 - accuracy: 0.5981 - val_loss: 0.7814 - val_accuracy: 0.6171\n",
      "Epoch 173/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8147 - accuracy: 0.5973 - val_loss: 0.7807 - val_accuracy: 0.6104\n",
      "Epoch 174/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8127 - accuracy: 0.5847 - val_loss: 0.7808 - val_accuracy: 0.6109\n",
      "Epoch 175/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8167 - accuracy: 0.5962 - val_loss: 0.7808 - val_accuracy: 0.6204\n",
      "Epoch 176/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8034 - accuracy: 0.6093 - val_loss: 0.7794 - val_accuracy: 0.6190\n",
      "Epoch 177/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8090 - accuracy: 0.5978 - val_loss: 0.7798 - val_accuracy: 0.6196\n",
      "Epoch 178/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8075 - accuracy: 0.5995 - val_loss: 0.7787 - val_accuracy: 0.6151\n",
      "Epoch 179/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8215 - accuracy: 0.5984 - val_loss: 0.7791 - val_accuracy: 0.6151\n",
      "Epoch 180/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8040 - accuracy: 0.6003 - val_loss: 0.7780 - val_accuracy: 0.6215\n",
      "Epoch 181/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8132 - accuracy: 0.5939 - val_loss: 0.7792 - val_accuracy: 0.6198\n",
      "Epoch 182/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8174 - accuracy: 0.5950 - val_loss: 0.7799 - val_accuracy: 0.6145\n",
      "Epoch 183/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8125 - accuracy: 0.5936 - val_loss: 0.7774 - val_accuracy: 0.6196\n",
      "Epoch 184/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8139 - accuracy: 0.5998 - val_loss: 0.7789 - val_accuracy: 0.6157\n",
      "Epoch 185/300\n",
      "57/57 [==============================] - 0s 4ms/step - loss: 0.8053 - accuracy: 0.5964 - val_loss: 0.7780 - val_accuracy: 0.6162\n",
      "Epoch 186/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8155 - accuracy: 0.6012 - val_loss: 0.7777 - val_accuracy: 0.6212\n",
      "Epoch 187/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8130 - accuracy: 0.5978 - val_loss: 0.7769 - val_accuracy: 0.6235\n",
      "Epoch 188/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8114 - accuracy: 0.6051 - val_loss: 0.7767 - val_accuracy: 0.6157\n",
      "Epoch 189/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8121 - accuracy: 0.6006 - val_loss: 0.7776 - val_accuracy: 0.6240\n",
      "Epoch 190/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8137 - accuracy: 0.5967 - val_loss: 0.7743 - val_accuracy: 0.6274\n",
      "Epoch 191/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8025 - accuracy: 0.6017 - val_loss: 0.7750 - val_accuracy: 0.6173\n",
      "Epoch 192/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8061 - accuracy: 0.6012 - val_loss: 0.7738 - val_accuracy: 0.6257\n",
      "Epoch 193/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8053 - accuracy: 0.5984 - val_loss: 0.7748 - val_accuracy: 0.6276\n",
      "Epoch 194/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8100 - accuracy: 0.5992 - val_loss: 0.7735 - val_accuracy: 0.6162\n",
      "Epoch 195/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8073 - accuracy: 0.5945 - val_loss: 0.7762 - val_accuracy: 0.6176\n",
      "Epoch 196/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8016 - accuracy: 0.5917 - val_loss: 0.7770 - val_accuracy: 0.6232\n",
      "Epoch 197/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8143 - accuracy: 0.5950 - val_loss: 0.7756 - val_accuracy: 0.6215\n",
      "Epoch 198/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8136 - accuracy: 0.5934 - val_loss: 0.7759 - val_accuracy: 0.6140\n",
      "Epoch 199/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8150 - accuracy: 0.5998 - val_loss: 0.7757 - val_accuracy: 0.6215\n",
      "Epoch 200/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8081 - accuracy: 0.5872 - val_loss: 0.7730 - val_accuracy: 0.6212\n",
      "Epoch 201/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8080 - accuracy: 0.6020 - val_loss: 0.7723 - val_accuracy: 0.6268\n",
      "Epoch 202/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8013 - accuracy: 0.5970 - val_loss: 0.7724 - val_accuracy: 0.6179\n",
      "Epoch 203/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8198 - accuracy: 0.6081 - val_loss: 0.7715 - val_accuracy: 0.6196\n",
      "Epoch 204/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8087 - accuracy: 0.5992 - val_loss: 0.7731 - val_accuracy: 0.6260\n",
      "Epoch 205/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8112 - accuracy: 0.6048 - val_loss: 0.7710 - val_accuracy: 0.6229\n",
      "Epoch 206/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8137 - accuracy: 0.5959 - val_loss: 0.7722 - val_accuracy: 0.6129\n",
      "Epoch 207/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7998 - accuracy: 0.5975 - val_loss: 0.7717 - val_accuracy: 0.6254\n",
      "Epoch 208/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7973 - accuracy: 0.5967 - val_loss: 0.7715 - val_accuracy: 0.6246\n",
      "Epoch 209/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8057 - accuracy: 0.5995 - val_loss: 0.7712 - val_accuracy: 0.6134\n",
      "Epoch 210/300\n",
      "57/57 [==============================] - 0s 4ms/step - loss: 0.8045 - accuracy: 0.5950 - val_loss: 0.7683 - val_accuracy: 0.6165\n",
      "Epoch 211/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8041 - accuracy: 0.6051 - val_loss: 0.7691 - val_accuracy: 0.6237\n",
      "Epoch 212/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8097 - accuracy: 0.5906 - val_loss: 0.7693 - val_accuracy: 0.6240\n",
      "Epoch 213/300\n",
      "57/57 [==============================] - 0s 4ms/step - loss: 0.8107 - accuracy: 0.5939 - val_loss: 0.7699 - val_accuracy: 0.6240\n",
      "Epoch 214/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8101 - accuracy: 0.5956 - val_loss: 0.7691 - val_accuracy: 0.6212\n",
      "Epoch 215/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7945 - accuracy: 0.6098 - val_loss: 0.7694 - val_accuracy: 0.6212\n",
      "Epoch 216/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8041 - accuracy: 0.5962 - val_loss: 0.7675 - val_accuracy: 0.6198\n",
      "Epoch 217/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8028 - accuracy: 0.6037 - val_loss: 0.7711 - val_accuracy: 0.6232\n",
      "Epoch 218/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8020 - accuracy: 0.6062 - val_loss: 0.7678 - val_accuracy: 0.6171\n",
      "Epoch 219/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8069 - accuracy: 0.6104 - val_loss: 0.7693 - val_accuracy: 0.6137\n",
      "Epoch 220/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7953 - accuracy: 0.6054 - val_loss: 0.7671 - val_accuracy: 0.6218\n",
      "Epoch 221/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8023 - accuracy: 0.5975 - val_loss: 0.7668 - val_accuracy: 0.6226\n",
      "Epoch 222/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8090 - accuracy: 0.5998 - val_loss: 0.7674 - val_accuracy: 0.6263\n",
      "Epoch 223/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7969 - accuracy: 0.6003 - val_loss: 0.7668 - val_accuracy: 0.6215\n",
      "Epoch 224/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7986 - accuracy: 0.6067 - val_loss: 0.7678 - val_accuracy: 0.6246\n",
      "Epoch 225/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8000 - accuracy: 0.6048 - val_loss: 0.7684 - val_accuracy: 0.6268\n",
      "Epoch 226/300\n",
      "57/57 [==============================] - 0s 4ms/step - loss: 0.7967 - accuracy: 0.6081 - val_loss: 0.7670 - val_accuracy: 0.6274\n",
      "Epoch 227/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8021 - accuracy: 0.6031 - val_loss: 0.7682 - val_accuracy: 0.6182\n",
      "Epoch 228/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7956 - accuracy: 0.5917 - val_loss: 0.7677 - val_accuracy: 0.6263\n",
      "Epoch 229/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8145 - accuracy: 0.5992 - val_loss: 0.7671 - val_accuracy: 0.6276\n",
      "Epoch 230/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7989 - accuracy: 0.5984 - val_loss: 0.7662 - val_accuracy: 0.6246\n",
      "Epoch 231/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8097 - accuracy: 0.6067 - val_loss: 0.7666 - val_accuracy: 0.6151\n",
      "Epoch 232/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8069 - accuracy: 0.5953 - val_loss: 0.7668 - val_accuracy: 0.6260\n",
      "Epoch 233/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8003 - accuracy: 0.6028 - val_loss: 0.7649 - val_accuracy: 0.6265\n",
      "Epoch 234/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7979 - accuracy: 0.5984 - val_loss: 0.7655 - val_accuracy: 0.6240\n",
      "Epoch 235/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7906 - accuracy: 0.6076 - val_loss: 0.7655 - val_accuracy: 0.6268\n",
      "Epoch 236/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8013 - accuracy: 0.6054 - val_loss: 0.7651 - val_accuracy: 0.6237\n",
      "Epoch 237/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7963 - accuracy: 0.6031 - val_loss: 0.7639 - val_accuracy: 0.6285\n",
      "Epoch 238/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8112 - accuracy: 0.6014 - val_loss: 0.7655 - val_accuracy: 0.6171\n",
      "Epoch 239/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7973 - accuracy: 0.6095 - val_loss: 0.7655 - val_accuracy: 0.6218\n",
      "Epoch 240/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7986 - accuracy: 0.5987 - val_loss: 0.7650 - val_accuracy: 0.6187\n",
      "Epoch 241/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8079 - accuracy: 0.5956 - val_loss: 0.7638 - val_accuracy: 0.6176\n",
      "Epoch 242/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7951 - accuracy: 0.6001 - val_loss: 0.7636 - val_accuracy: 0.6196\n",
      "Epoch 243/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7953 - accuracy: 0.6009 - val_loss: 0.7654 - val_accuracy: 0.6212\n",
      "Epoch 244/300\n",
      "57/57 [==============================] - 0s 4ms/step - loss: 0.7980 - accuracy: 0.6054 - val_loss: 0.7633 - val_accuracy: 0.6268\n",
      "Epoch 245/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7928 - accuracy: 0.5967 - val_loss: 0.7647 - val_accuracy: 0.6257\n",
      "Epoch 246/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8019 - accuracy: 0.6126 - val_loss: 0.7649 - val_accuracy: 0.6240\n",
      "Epoch 247/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8025 - accuracy: 0.6020 - val_loss: 0.7632 - val_accuracy: 0.6137\n",
      "Epoch 248/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8005 - accuracy: 0.5959 - val_loss: 0.7639 - val_accuracy: 0.6235\n",
      "Epoch 249/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8003 - accuracy: 0.5956 - val_loss: 0.7628 - val_accuracy: 0.6251\n",
      "Epoch 250/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7907 - accuracy: 0.6045 - val_loss: 0.7642 - val_accuracy: 0.6171\n",
      "Epoch 251/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8071 - accuracy: 0.6009 - val_loss: 0.7638 - val_accuracy: 0.6237\n",
      "Epoch 252/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8054 - accuracy: 0.5975 - val_loss: 0.7632 - val_accuracy: 0.6162\n",
      "Epoch 253/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7978 - accuracy: 0.6034 - val_loss: 0.7621 - val_accuracy: 0.6235\n",
      "Epoch 254/300\n",
      "57/57 [==============================] - 0s 4ms/step - loss: 0.7938 - accuracy: 0.6020 - val_loss: 0.7609 - val_accuracy: 0.6235\n",
      "Epoch 255/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7931 - accuracy: 0.6051 - val_loss: 0.7605 - val_accuracy: 0.6279\n",
      "Epoch 256/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7961 - accuracy: 0.6067 - val_loss: 0.7608 - val_accuracy: 0.6257\n",
      "Epoch 257/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8049 - accuracy: 0.5964 - val_loss: 0.7621 - val_accuracy: 0.6293\n",
      "Epoch 258/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7998 - accuracy: 0.5967 - val_loss: 0.7601 - val_accuracy: 0.6260\n",
      "Epoch 259/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8103 - accuracy: 0.5903 - val_loss: 0.7626 - val_accuracy: 0.6224\n",
      "Epoch 260/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7997 - accuracy: 0.5892 - val_loss: 0.7601 - val_accuracy: 0.6274\n",
      "Epoch 261/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8035 - accuracy: 0.5959 - val_loss: 0.7610 - val_accuracy: 0.6288\n",
      "Epoch 262/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7900 - accuracy: 0.6042 - val_loss: 0.7600 - val_accuracy: 0.6249\n",
      "Epoch 263/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7958 - accuracy: 0.5973 - val_loss: 0.7626 - val_accuracy: 0.6176\n",
      "Epoch 264/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7923 - accuracy: 0.6056 - val_loss: 0.7596 - val_accuracy: 0.6176\n",
      "Epoch 265/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7931 - accuracy: 0.5995 - val_loss: 0.7597 - val_accuracy: 0.6290\n",
      "Epoch 266/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8000 - accuracy: 0.5967 - val_loss: 0.7594 - val_accuracy: 0.6263\n",
      "Epoch 267/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7998 - accuracy: 0.6003 - val_loss: 0.7591 - val_accuracy: 0.6274\n",
      "Epoch 268/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7865 - accuracy: 0.5950 - val_loss: 0.7604 - val_accuracy: 0.6232\n",
      "Epoch 269/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7869 - accuracy: 0.5967 - val_loss: 0.7585 - val_accuracy: 0.6274\n",
      "Epoch 270/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8065 - accuracy: 0.5989 - val_loss: 0.7596 - val_accuracy: 0.6268\n",
      "Epoch 271/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7899 - accuracy: 0.6054 - val_loss: 0.7602 - val_accuracy: 0.6198\n",
      "Epoch 272/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7882 - accuracy: 0.5959 - val_loss: 0.7605 - val_accuracy: 0.6240\n",
      "Epoch 273/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7982 - accuracy: 0.5989 - val_loss: 0.7610 - val_accuracy: 0.6218\n",
      "Epoch 274/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7978 - accuracy: 0.5973 - val_loss: 0.7582 - val_accuracy: 0.6235\n",
      "Epoch 275/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8023 - accuracy: 0.6090 - val_loss: 0.7587 - val_accuracy: 0.6274\n",
      "Epoch 276/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8016 - accuracy: 0.6028 - val_loss: 0.7584 - val_accuracy: 0.6243\n",
      "Epoch 277/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7960 - accuracy: 0.6028 - val_loss: 0.7608 - val_accuracy: 0.6279\n",
      "Epoch 278/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7914 - accuracy: 0.6051 - val_loss: 0.7598 - val_accuracy: 0.6257\n",
      "Epoch 279/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8004 - accuracy: 0.6014 - val_loss: 0.7603 - val_accuracy: 0.6271\n",
      "Epoch 280/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7905 - accuracy: 0.5998 - val_loss: 0.7574 - val_accuracy: 0.6302\n",
      "Epoch 281/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7928 - accuracy: 0.6073 - val_loss: 0.7587 - val_accuracy: 0.6212\n",
      "Epoch 282/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7905 - accuracy: 0.5998 - val_loss: 0.7572 - val_accuracy: 0.6302\n",
      "Epoch 283/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7908 - accuracy: 0.6020 - val_loss: 0.7566 - val_accuracy: 0.6218\n",
      "Epoch 284/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7990 - accuracy: 0.5981 - val_loss: 0.7577 - val_accuracy: 0.6257\n",
      "Epoch 285/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7914 - accuracy: 0.6003 - val_loss: 0.7570 - val_accuracy: 0.6282\n",
      "Epoch 286/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.8065 - accuracy: 0.6090 - val_loss: 0.7579 - val_accuracy: 0.6263\n",
      "Epoch 287/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8002 - accuracy: 0.6006 - val_loss: 0.7558 - val_accuracy: 0.6260\n",
      "Epoch 288/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8018 - accuracy: 0.6084 - val_loss: 0.7568 - val_accuracy: 0.6221\n",
      "Epoch 289/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7853 - accuracy: 0.6093 - val_loss: 0.7558 - val_accuracy: 0.6201\n",
      "Epoch 290/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7958 - accuracy: 0.5987 - val_loss: 0.7566 - val_accuracy: 0.6249\n",
      "Epoch 291/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7838 - accuracy: 0.6187 - val_loss: 0.7589 - val_accuracy: 0.6271\n",
      "Epoch 292/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7923 - accuracy: 0.6140 - val_loss: 0.7572 - val_accuracy: 0.6260\n",
      "Epoch 293/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7894 - accuracy: 0.6070 - val_loss: 0.7565 - val_accuracy: 0.6260\n",
      "Epoch 294/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.8020 - accuracy: 0.6106 - val_loss: 0.7566 - val_accuracy: 0.6215\n",
      "Epoch 295/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7950 - accuracy: 0.6093 - val_loss: 0.7540 - val_accuracy: 0.6265\n",
      "Epoch 296/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7788 - accuracy: 0.6087 - val_loss: 0.7549 - val_accuracy: 0.6293\n",
      "Epoch 297/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7861 - accuracy: 0.6081 - val_loss: 0.7544 - val_accuracy: 0.6268\n",
      "Epoch 298/300\n",
      "57/57 [==============================] - 0s 2ms/step - loss: 0.7909 - accuracy: 0.6020 - val_loss: 0.7559 - val_accuracy: 0.6232\n",
      "Epoch 299/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7860 - accuracy: 0.6065 - val_loss: 0.7551 - val_accuracy: 0.6285\n",
      "Epoch 300/300\n",
      "57/57 [==============================] - 0s 3ms/step - loss: 0.7933 - accuracy: 0.5981 - val_loss: 0.7538 - val_accuracy: 0.6235\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.src.callbacks.History at 0x7fba48efbb80>"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(\n",
    "    x=X_train, \n",
    "    y=y_train_cat, \n",
    "    epochs=300,\n",
    "    validation_data=(X_test, y_test_cat),\n",
    "    verbose=1,\n",
    "    batch_size=64,\n",
    "    callbacks=[early_stop]\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2c0eb188-26ed-420f-9674-3ee1c4b44156",
   "metadata": {},
   "source": [
    "Let's plot the Loss vs Epoch chart."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "2cae56b6-8e24-47c3-9f2e-dda2538eed56",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model_loss = pd.DataFrame(model.history.history)\n",
    "model_loss[['loss','val_loss']].plot()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bec4469c-aef3-46c0-bb0f-c4680ae87744",
   "metadata": {},
   "source": [
    "Let's evaluate the trained model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "6dc6c2e8-0fb7-4e31-8352-ea82effcd7f5",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "113/113 [==============================] - 0s 1ms/step - loss: 0.7538 - accuracy: 0.6235\n",
      "Loss, Accuracy:  [0.7537979483604431, 0.6234670877456665]\n"
     ]
    }
   ],
   "source": [
    "print(\"Loss, Accuracy: \", model.evaluate(X_test, y_test_cat))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "f9953c95-0bb6-4d6f-9444-a9c7a83af1df",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "113/113 [==============================] - 0s 1ms/step\n",
      "F1-score:  0.6167088942743223\n"
     ]
    }
   ],
   "source": [
    "print(\"F1-score: \", f1_score(y_test, np.argmax(model.predict(X_test), axis=1), average='weighted'))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ccebd798-36ee-4e3b-a364-6e688a890c67",
   "metadata": {},
   "source": [
    "With a simple, unoptimized NN model a F1-score of approximately 62% was achieved.\n",
    "\n",
    "There is always room for model fine-tuning, optimizations and futher model improvements."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.14"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}